File size: 18,307 Bytes
4c65bff
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
# coding=utf-8
# Copyright 2021 The HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
 Sequence feature extraction class for common feature extractors to preprocess sequences.
"""
from typing import Dict, List, Optional, Union

import numpy as np

from .feature_extraction_utils import BatchFeature, FeatureExtractionMixin
from .utils import PaddingStrategy, TensorType, is_tf_tensor, is_torch_tensor, logging, to_numpy


logger = logging.get_logger(__name__)


class SequenceFeatureExtractor(FeatureExtractionMixin):
    """
    This is a general feature extraction class for speech recognition.

    Args:
        feature_size (`int`):
            The feature dimension of the extracted features.
        sampling_rate (`int`):
            The sampling rate at which the audio files should be digitalized expressed in hertz (Hz).
        padding_value (`float`):
            The value that is used to fill the padding values / vectors.
    """

    def __init__(self, feature_size: int, sampling_rate: int, padding_value: float, **kwargs):
        self.feature_size = feature_size
        self.sampling_rate = sampling_rate
        self.padding_value = padding_value

        self.padding_side = kwargs.pop("padding_side", "right")
        self.return_attention_mask = kwargs.pop("return_attention_mask", True)

        super().__init__(**kwargs)

    def pad(
        self,
        processed_features: Union[
            BatchFeature,
            List[BatchFeature],
            Dict[str, BatchFeature],
            Dict[str, List[BatchFeature]],
            List[Dict[str, BatchFeature]],
        ],
        padding: Union[bool, str, PaddingStrategy] = True,
        max_length: Optional[int] = None,
        truncation: bool = False,
        pad_to_multiple_of: Optional[int] = None,
        return_attention_mask: Optional[bool] = None,
        return_tensors: Optional[Union[str, TensorType]] = None,
    ) -> BatchFeature:
        """
        Pad input values / input vectors or a batch of input values / input vectors up to predefined length or to the
        max sequence length in the batch.

        Padding side (left/right) padding values are defined at the feature extractor level (with `self.padding_side`,
        `self.padding_value`)

        <Tip>

        If the `processed_features` passed are dictionary of numpy arrays, PyTorch tensors or TensorFlow tensors, the
        result will use the same type unless you provide a different tensor type with `return_tensors`. In the case of
        PyTorch tensors, you will lose the specific device of your tensors however.

        </Tip>

        Args:
            processed_features ([`BatchFeature`], list of [`BatchFeature`], `Dict[str, List[float]]`, `Dict[str, List[List[float]]` or `List[Dict[str, List[float]]]`):
                Processed inputs. Can represent one input ([`BatchFeature`] or `Dict[str, List[float]]`) or a batch of
                input values / vectors (list of [`BatchFeature`], *Dict[str, List[List[float]]]* or *List[Dict[str,
                List[float]]]*) so you can use this method during preprocessing as well as in a PyTorch Dataloader
                collate function.

                Instead of `List[float]` you can have tensors (numpy arrays, PyTorch tensors or TensorFlow tensors),
                see the note above for the return type.
            padding (`bool`, `str` or [`~utils.PaddingStrategy`], *optional*, defaults to `True`):
                Select a strategy to pad the returned sequences (according to the model's padding side and padding
                index) among:

                - `True` or `'longest'`: Pad to the longest sequence in the batch (or no padding if only a single
                  sequence if provided).
                - `'max_length'`: Pad to a maximum length specified with the argument `max_length` or to the maximum
                  acceptable input length for the model if that argument is not provided.
                - `False` or `'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of different
                  lengths).
            max_length (`int`, *optional*):
                Maximum length of the returned list and optionally padding length (see above).
            truncation (`bool`):
                Activates truncation to cut input sequences longer than `max_length` to `max_length`.
            pad_to_multiple_of (`int`, *optional*):
                If set will pad the sequence to a multiple of the provided value.

                This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability
                `>= 7.5` (Volta), or on TPUs which benefit from having sequence lengths be a multiple of 128.
            return_attention_mask (`bool`, *optional*):
                Whether to return the attention mask. If left to the default, will return the attention mask according
                to the specific feature_extractor's default.

                [What are attention masks?](../glossary#attention-mask)
            return_tensors (`str` or [`~utils.TensorType`], *optional*):
                If set, will return tensors instead of list of python integers. Acceptable values are:

                - `'tf'`: Return TensorFlow `tf.constant` objects.
                - `'pt'`: Return PyTorch `torch.Tensor` objects.
                - `'np'`: Return Numpy `np.ndarray` objects.
        """
        # If we have a list of dicts, let's convert it in a dict of lists
        # We do this to allow using this method as a collate_fn function in PyTorch Dataloader
        if isinstance(processed_features, (list, tuple)) and isinstance(processed_features[0], (dict, BatchFeature)):
            processed_features = {
                key: [example[key] for example in processed_features] for key in processed_features[0].keys()
            }

        # The model's main input name, usually `input_values`, has be passed for padding
        if self.model_input_names[0] not in processed_features:
            raise ValueError(
                "You should supply an instance of `transformers.BatchFeature` or list of `transformers.BatchFeature`"
                f" to this method that includes {self.model_input_names[0]}, but you provided"
                f" {list(processed_features.keys())}"
            )

        required_input = processed_features[self.model_input_names[0]]
        return_attention_mask = (
            return_attention_mask if return_attention_mask is not None else self.return_attention_mask
        )

        if len(required_input) == 0:
            if return_attention_mask:
                processed_features["attention_mask"] = []
            return processed_features

        # If we have PyTorch/TF tensors or lists as inputs, we cast them as Numpy arrays
        # and rebuild them afterwards if no return_tensors is specified
        # Note that we lose the specific device the tensor may be on for PyTorch

        first_element = required_input[0]
        if isinstance(first_element, (list, tuple)):
            # first_element might be an empty list/tuple in some edge cases so we grab the first non empty element.
            index = 0
            while len(required_input[index]) == 0:
                index += 1
            if index < len(required_input):
                first_element = required_input[index][0]

        if return_tensors is None:
            if is_tf_tensor(first_element):
                return_tensors = "tf"
            elif is_torch_tensor(first_element):
                return_tensors = "pt"
            elif isinstance(first_element, (int, float, list, tuple, np.ndarray)):
                return_tensors = "np"
            else:
                raise ValueError(
                    f"type of {first_element} unknown: {type(first_element)}. "
                    "Should be one of a python, numpy, pytorch or tensorflow object."
                )

        for key, value in processed_features.items():
            if isinstance(value[0], (int, float)):
                processed_features[key] = to_numpy(value)
            else:
                processed_features[key] = [to_numpy(v) for v in value]

        # Convert padding_strategy in PaddingStrategy
        padding_strategy = self._get_padding_strategies(padding=padding, max_length=max_length)

        required_input = processed_features[self.model_input_names[0]]

        batch_size = len(required_input)
        if not all(len(v) == batch_size for v in processed_features.values()):
            raise ValueError("Some items in the output dictionary have a different batch size than others.")

        truncated_inputs = []
        for i in range(batch_size):
            inputs = {k: v[i] for k, v in processed_features.items()}
            # truncation
            inputs_slice = self._truncate(
                inputs,
                max_length=max_length,
                pad_to_multiple_of=pad_to_multiple_of,
                truncation=truncation,
            )
            truncated_inputs.append(inputs_slice)

        if padding_strategy == PaddingStrategy.LONGEST:
            # make sure that `max_length` cannot be longer than the longest truncated length
            max_length = max(len(input_slice[self.model_input_names[0]]) for input_slice in truncated_inputs)
            padding_strategy = PaddingStrategy.MAX_LENGTH

        batch_outputs = {}
        for i in range(batch_size):
            # padding
            outputs = self._pad(
                truncated_inputs[i],
                max_length=max_length,
                padding_strategy=padding_strategy,
                pad_to_multiple_of=pad_to_multiple_of,
                return_attention_mask=return_attention_mask,
            )

            for key, value in outputs.items():
                if key not in batch_outputs:
                    batch_outputs[key] = []
                if value.dtype is np.dtype(np.float64):
                    value = value.astype(np.float32)
                batch_outputs[key].append(value)

        return BatchFeature(batch_outputs, tensor_type=return_tensors)

    def _pad(
        self,
        processed_features: Union[Dict[str, np.ndarray], BatchFeature],
        max_length: Optional[int] = None,
        padding_strategy: PaddingStrategy = PaddingStrategy.DO_NOT_PAD,
        pad_to_multiple_of: Optional[int] = None,
        return_attention_mask: Optional[bool] = None,
    ) -> dict:
        """
        Pad inputs (on left/right and up to predefined length or max length in the batch)

        Args:
            processed_features (`Union[Dict[str, np.ndarray], BatchFeature]`):
                Dictionary of input values (`np.ndarray[float]`) / input vectors (`List[np.ndarray[float]]`) or batch
                of inputs values (`List[np.ndarray[int]]`) / input vectors (`List[np.ndarray[int]]`)
            max_length (`int`, *optional*):
                Maximum length of the returned list and optionally padding length (see below)
            padding_strategy (`PaddingStrategy`, *optional*, default to `PaddingStrategy.DO_NOT_PAD`):
                PaddingStrategy to use for padding.

                - PaddingStrategy.LONGEST Pad to the longest sequence in the batch
                - PaddingStrategy.MAX_LENGTH: Pad to the max length (default)
                - PaddingStrategy.DO_NOT_PAD: Do not pad
                The feature_extractor padding sides are defined in self.padding_side:

                    - 'left': pads on the left of the sequences
                    - 'right': pads on the right of the sequences
            pad_to_multiple_of (`int`, *optional*):
                Integer if set will pad the sequence to a multiple of the provided value. This is especially useful to
                enable the use of Tensor Core on NVIDIA hardware with compute capability `>= 7.5` (Volta), or on TPUs
                which benefit from having sequence lengths be a multiple of 128.
            return_attention_mask (`bool`, *optional*):
                Set to False to avoid returning attention mask (default: set to model specifics)
        """
        required_input = processed_features[self.model_input_names[0]]

        if padding_strategy == PaddingStrategy.LONGEST:
            max_length = len(required_input)

        if max_length is not None and pad_to_multiple_of is not None and (max_length % pad_to_multiple_of != 0):
            max_length = ((max_length // pad_to_multiple_of) + 1) * pad_to_multiple_of

        needs_to_be_padded = padding_strategy != PaddingStrategy.DO_NOT_PAD and len(required_input) < max_length

        if return_attention_mask and "attention_mask" not in processed_features:
            processed_features["attention_mask"] = np.ones(len(required_input), dtype=np.int32)

        if needs_to_be_padded:
            difference = max_length - len(required_input)
            if self.padding_side == "right":
                if return_attention_mask:
                    processed_features["attention_mask"] = np.pad(
                        processed_features["attention_mask"], (0, difference)
                    )
                padding_shape = ((0, difference), (0, 0)) if self.feature_size > 1 else (0, difference)
                processed_features[self.model_input_names[0]] = np.pad(
                    required_input, padding_shape, "constant", constant_values=self.padding_value
                )
            elif self.padding_side == "left":
                if return_attention_mask:
                    processed_features["attention_mask"] = np.pad(
                        processed_features["attention_mask"], (difference, 0)
                    )
                padding_shape = ((difference, 0), (0, 0)) if self.feature_size > 1 else (difference, 0)
                processed_features[self.model_input_names[0]] = np.pad(
                    required_input, padding_shape, "constant", constant_values=self.padding_value
                )
            else:
                raise ValueError("Invalid padding strategy:" + str(self.padding_side))

        return processed_features

    def _truncate(
        self,
        processed_features: Union[Dict[str, np.ndarray], BatchFeature],
        max_length: Optional[int] = None,
        pad_to_multiple_of: Optional[int] = None,
        truncation: Optional[bool] = None,
    ):
        """
        Truncate inputs to predefined length or max length in the batch

        Args:
            processed_features(`Union[Dict[str, np.ndarray], BatchFeature]`):
                Dictionary of input values (`np.ndarray[float]`) / input vectors (`List[np.ndarray[float]]`) or batch
                of inputs values (`List[np.ndarray[int]]`) / input vectors (`List[np.ndarray[int]]`)
            max_length (`int`, *optional*):
                maximum length of the returned list and optionally padding length (see below)
            pad_to_multiple_of (`int`, *optional*) :
                Integer if set will pad the sequence to a multiple of the provided value. This is especially useful to
                enable the use of Tensor Core on NVIDIA hardware with compute capability `>= 7.5` (Volta), or on TPUs
                which benefit from having sequence lengths be a multiple of 128.
            truncation (`bool`, *optional*):
                Activates truncation to cut input sequences longer than `max_length` to `max_length`.
        """
        if not truncation:
            return processed_features
        elif truncation and max_length is None:
            raise ValueError("When setting ``truncation=True``, make sure that ``max_length`` is defined.")

        required_input = processed_features[self.model_input_names[0]]

        # find `max_length` that fits `pad_to_multiple_of`
        if max_length is not None and pad_to_multiple_of is not None and (max_length % pad_to_multiple_of != 0):
            max_length = ((max_length // pad_to_multiple_of) + 1) * pad_to_multiple_of

        needs_to_be_truncated = len(required_input) > max_length

        if needs_to_be_truncated:
            processed_features[self.model_input_names[0]] = processed_features[self.model_input_names[0]][:max_length]
            if "attention_mask" in processed_features:
                processed_features["attention_mask"] = processed_features["attention_mask"][:max_length]

        return processed_features

    def _get_padding_strategies(self, padding=False, max_length=None):
        """
        Find the correct padding strategy
        """

        # Get padding strategy
        if padding is not False:
            if padding is True:
                padding_strategy = PaddingStrategy.LONGEST  # Default to pad to the longest sequence in the batch
            elif not isinstance(padding, PaddingStrategy):
                padding_strategy = PaddingStrategy(padding)
            elif isinstance(padding, PaddingStrategy):
                padding_strategy = padding
        else:
            padding_strategy = PaddingStrategy.DO_NOT_PAD

        # Set max length if needed
        if max_length is None:
            if padding_strategy == PaddingStrategy.MAX_LENGTH:
                raise ValueError(
                    f"When setting ``padding={PaddingStrategy.MAX_LENGTH}``, make sure that max_length is defined"
                )

        # Test if we have a padding value
        if padding_strategy != PaddingStrategy.DO_NOT_PAD and (self.padding_value is None):
            raise ValueError(
                "Asking to pad but the feature_extractor does not have a padding value. Please select a value to use"
                " as `padding_value`. For example: `feature_extractor.padding_value = 0.0`."
            )

        return padding_strategy